""PrecisionMachinisT"" wrote in message
...

Predictive correlations are the basis of much of the marketing
statistics
that are in use today. Whether one event causes the other hardly
matters --
likely there is some unknown, third factor which is the causative one
for
both of the events you're tracking -- but the ability of the data to
predict
is its value. It starts with a correlation like the one you've
described.


Perhaps a poor choice of words on my part...........

I was actually going to introduce a third set of data into the landscape (
sudden export restrictions on gold from the Soviet Union or somesuch )

And suppose upon noting a change occurred in the previous correlation
between the price of tea in China, and data concerning deadly automobile
accidents in the USA; question whether it could rightly be assumed to have
been caused by the said gold export restrictions...........

Of course, still, the answer is no. g

But I ran out of time, and so I didnt elaborate nor edit the previous post
much for clarity sake.

The important part being you did understand exactly what it was I was
*trying* to say, regardless of my perhaps poor choice of terms.

Sure, and your point, although a little rough around the edges, is
essentially correct. Attributing causative relationships to mere
coincidences is one of the ways statistics are misunderstood, and a way
they're sometimes intentionally misused.

This is a difficult example, though, because it requires some understanding
of when correlations become meaningful. A couple of weeks ago I quoted here
from a statistics consultant about exactly this subject, in which he
described the common kind of after-the-fact regression analysis as junk
science. Data mining as it's practiced today by unsophisticated marketers
and by political parties often is that kind of junk science. It stops being
junk science when the correlation actually predicts events as well as
describing historical relationships. It doesn't have to show the pattern of
causation in order to be useful, but it does have to accurately predict
events, or it's just junk.

There are many ways in which statistics are misused and abused, many of them
simpler to follow than your example. The Interpol case is a good one. What
they've compiled is data that allows you to see what's happening to crime
within individual countries over time, but not much else. Because one
country counts only apples, while another counts only oranges, it's useless
for comparing fruit from one country with fruit from another. The statistics
are basically good for their intended purpose, in other words, but complete
junk when someone tries to make a different and improper use of the data.

--
Ed Huntress
(remove "3" from email address for email reply)